Building a Landing Page With AI Agents in a ClawChat Group

Zac gave two AI agents a real client brief in a ClawChat group. One built the landing page. The other reviewed the design and wrote the copy. They discussed the work with each other, and Zac answered two questions along the way.

Then a third agent reviewed the result while his laptop was asleep.

ClawChat sponsored the video. The test includes the finished page, the agents’ disagreements, and a decision they should have questioned but initially approved.

Watch Zac’s full workflow on YouTube.

Dev builds, Riley reviews

Zac started with two agents running locally on his Mac, each with its own ClawChat account.

Dev was the builder, running through OpenClaw and responsible for writing the code. Riley handled design and copy review. Zac had explicitly told Riley to push back when the work needed improvement.

He created a group called “client project,” added both agents, and posted the brief. Their roles were already established, but he did not break the brief into individual task assignments.

Dev proposed the page structure. Riley challenged parts of the plan, including the proposed example listings and form submission approach. Zac supplied the information they needed, including three real listings.

The resulting page included a hero section, value cards, featured tools, and a submission form.

The useful part of this workflow was seeing the implementation and review happen in the same conversation. Zac could follow the discussion and answer questions without relaying every message between separate chats.

The instruction neither agent challenged

Zac deliberately asked for the signup form at the bottom of the page. In his assessment, that placement worked against the page’s main goal of generating signups.

Dev followed the instruction. Riley approved the result.

This was a more revealing limitation than a reviewer simply agreeing with everything. Riley had already challenged other decisions. It still missed this one.

A working page and an agent’s approval did not settle whether the page served the client’s goal. Zac still needed to judge that himself.

A cloud agent reviews while the laptop sleeps

Next, Zac gave Ops, his cloud-hosted agent, a separate task: review the landing page’s copy and structure and return three concrete improvements.

He closed his laptop. When he reopened it, the review was ready.

Ops flagged the signup section’s placement and navigation, copy problems, and a featured-tools section that weakened the pitch.

This demonstrated a practical reason to combine local and cloud agents. The local agents handled the build while the Mac was available. Ops could continue its assigned review because it ran independently in the cloud.

It does not mean the local agents kept running after the laptop went to sleep.

Try it with a small project

Zac’s workflow shows how a shared chat can support agents with different responsibilities: building, reviewing, and bringing another perspective to the result.

It also shows why human approval still matters. Agents can question some decisions and overlook others, even when they have been told to push back.

Zac noted that connecting all three agents took some fiddling. Once connected, though, he could give them a brief, answer their questions, and review the output in one conversation.

Already running your own agents? Try ClawChat, connect an agent, and start with a small task you can evaluate yourself.